Optimizing inventory management incorporating delivery demand uncertainty at Turff

The research is conducted at a Dutch start-up Turff that provides student houses with tablets for tracking finances and consumption of shared goods, as well as makes fast-scheduled drink deliveries. The aim is to answer the following research question: How can inventory management be improved at Turff for their best selling delivery product, Heineken, in order to increase forecasting accuracy from about 75-78% to 85-90% and optimize inventory levels? Heineken accounts for 75-80% of all delivery sales of Turff, highlighting the importance of an individual inventory plan. Currently, the replenishment decisions are made with experience based calculations. This often leads to non-optimal inventory levels, time wasted on manual decision making, and dependency on one employee. To develop a more accurate forecast for the demand with event-driven seasonal characteristics, we test several different methods. We obtain an accuracy of about 87.9% with a model that’s properties well match the data by reflecting the past patterns into the future as well as incorporating the effect of independent variables affecting the demand. Based on the forecast, we develop a dynamic ordering policy to directly assist with weekly decision making.